As a team of four Canadian medical students, we came into TreeHacks with a desire to address some of the biggest challenges faced within the healthcare system, specifically in Emergency Medical Services (EMS). We came across a pilot project happening in rural Alberta, called the Alberta Stroke Ambulance (https://www.albertahealthservices.ca/services/Page11617.aspx). The program is designed to dispatch specialized ambulances to smoke patients, and initiate the process of treatment as soon as the patient is in the ambulance. As a result, the timing of the ambulance response becomes very important and directly affects the patients' chance of survival, and resulting quality of life.
What it does
Ready Response is an application that is used by paramedics. The input is the current location (i.e. they are done with a recent call, and are currently awaiting their next call). Using Machine Learning, the application uses factors such as the location of call hot-spots, and the severity of the calls, in order to predict the best location for the ambulance to await the next call, in order to minimize the distance to the next call, and ultimately improve public safety by utilizing intelligent emergency medical responses dispatches.
How we built it
Ready Responder uses an evolutionary machine learning algorithm in the Python language, with a learning data set provided from data extracted from 30'000 EMS calls in Boulder, Colorado. The mobile app was produced on Swift. Arcgis software was used for geospatial analysis of existing ambulance respond times, and for the display of pre- and post- Ready Response times on a neighbourhood-by-neighbourhood basis.
Challenges we ran into
Our biggest challenge was the effective implementation of the evolutionary machine learning algorithm, specifically in having the sufficient number of factors to generate a truly optimized location for inter-call ambulance stops. Our other challenge was the lack of accessibility to ambulance data in many jursidictions, especially within Canada, due to the stringent data regulations especially in healthcare. This showed us the importance of open-data policies implemented by local and national governments, as we were able to use the data from Boulder, CO as the ML learning data set.
Accomplishments that we're proud of
We're happy to have a working interface, and knowing that there is already an interest from the potential stakeholders. Our greatest potential achievement would be the implementation of the app, within a pilot project (i.e. the Alberta Stroke Ambulance) and seeing the direct affect on patient outcomes and observing the emotion and economic benefits that it'll bring.
What we learned
First of all, we learned to never assume things are ready, until they are, especially when it comes to the efficacy of an ML algorithm.
What's next for Ready Responder
We want to fix the current glitches, improve the interface, and share it with some the individuals we've had discussions with (i.e. the Alberta Stroke Ambulance project and the Ontario Paramedics Association). We hope to get their feedback on the prototype, before proceeding with improvement. We'd also want to inquire the possibility of getting further datasets from these partners, in order to increase the efficacy of the produce, especially within the contexts that it is to be deployed (i.e. within that particular city/province/state).